On Sat, 23 Apr 2016 19:22:04 -0600 Scott Robison <scott at casaderobison.com> wrote:
> So if you could make your table up of integers, floats, and text > with character limits on them you could get fixed-length rows, which > might reduce your access time by 60% or more. Such a decrease in > access time could mean the difference between being able to update a > database live or being able to update only during an overnight run. As I tried to make clear in my reply to Keith, efficiency concerns are beside the point. The theory underpinning SQL rests on predicate logic and set theory. Freeing the user from concerns of physical storage and addressing were objectives Codd cited in his first paper. In point of fact, unbounded-length records have been supported for decades. Sybase called them TEXT and IMAGE types. You could not search them, though, only retrieve them once the row had been located by other means. Why? Just as you posit: for efficiency. The physical row kept a "pointer" to the TEXT data and, yes, every physical row had the same length, for efficiency reasons. --jkl
